datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
openvid-wantrack-tracks
OpenVid-WanTrack — 121-frame tracks (tar shards)
121-frame filtered tracks from OpenVid used to build the TrackWan training set (~259k npzs, ~666 GB).
Format: packed into tracks-XXXXX.tar shards (~1,000 npz per shard) for efficient hosting.
Each file is <video_id>_<clip>_<start>to<end>.npz. The basename joins 1:1 with the
tracks and processed datasets. Consume by streaming the tars (WebDataset) or tar -xf.
Apache-2.0; derived from OpenVid (respect upstream terms).
openvid-wantrack-processed-v2
OpenVid-WanTrack Processed (v2, 720p, bf16)
FastVideo preprocessing parquets for training the TrackWan point-track-conditioned I2V model on
the OpenVid-derived WanTrack set. Each row is one 121-frame clip with its VAE latents, text and
image conditioning, and dense CoTracker3 tracks — everything the trainer memory-maps, so no video
decoding happens at train time.
This is the bfloat16 variant of …/openvid-wantrack-processed (v2, 720p): the large float
tensor fields are stored in… See the full description on the dataset page: https://huggingface.co/datasets/FastVideo/openvid-wantrack-processed-v2.wantrack-synth-toy-720p
WanTrack synth toy — 720p
A 720p/24fps recreation of noctuashap/wantrack_synth_toy:
one shared seed image + 50 motion captions → 50 I2V clips, each varying only the motion.
Used as the overfit set for the bidirectional TrackWan teacher recipe in
FastVideo.
How it was built
Generation — Wan2.1-I2V-14B-720P, conditioned on the single synthetic_seed.png + each
line of captions.txt, 720×1280, 121 frames @ 24fps (gen_synth_i2v_worker.py).
Tracks — CoTracker3 on a… See the full description on the dataset page: https://huggingface.co/datasets/FastVideo/wantrack-synth-toy-720p.openvid-wantrack-processed
OpenVid-WanTrack — preprocessed training data
Preprocessed (FastVideo parquet) data used to train the TrackWan point-track-conditioned
video model. Derived from OpenVid; each row is one 121-frame clip with everything the
trainer consumes, so it is self-contained (no re-encoding needed).
Contents (parquet columns)
Each field is stored as *_bytes / *_shape / *_dtype (numpy buffers):
vae_latent — Wan VAE latent of the clip
first_frame_latent — I2V conditioning… See the full description on the dataset page: https://huggingface.co/datasets/noctuashap/openvid-wantrack-processed.wantrack_synth_toy
wantrack_synth_toy
20 synthetic Wan2.2-14B-I2V videos generated from a single shared first frame, with CoTracker3 tracks,
FastSAM object_ids, and lowrank track_weights baked in. Used for WanTrack (MotionStream-style) sparse-track
overfit experiments.
Layout
synthetic_seed.png -- shared first frame used by all 20 I2V generations.
videos/vid_NNNNNN.mp4 -- 20 clips, 121 frames, 480x832 @ 24 fps.
tracks/vid_NNNNNN.npz -- CoTracker3 50x50 grid tracks (tracks… See the full description on the dataset page: https://huggingface.co/datasets/noctuashap/wantrack_synth_toy.openvid-wantrack-clips
OpenVid-WanTrack — 121-frame clips (tar shards)
121-frame filtered clips from OpenVid used to build the TrackWan training set (~259k mp4s, ~799 GB).
Format: packed into clips-XXXXX.tar shards (~1,000 mp4 per shard) for efficient hosting.
Each file is <video_id>_<clip>_<start>to<end>.mp4. The basename joins 1:1 with the
tracks and processed datasets. Consume by streaming the tars (WebDataset) or tar -xf.
Apache-2.0; derived from OpenVid (respect upstream terms).